Professional Documents
Culture Documents
_
_
_
(3)
Similar expressions hold for the [C] and [K] matrices and
the load vector F . In Eq. (3), the zeros represent null
matrices of the appropriate dimensions. Subscripts i and
r denote internal and retained degrees of freedom, re-
spectively, and superscripts denote the substructure num-
ber or the transpose of a matrix.
The internal displacement vector U
`
i
for the `th
substructure is expressed as
U
`
i
= [X
`
[u
`
[T
`
[U
r
(4)
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2011
The transformation matrix [X
`
[ consists of a set of M-
orthonormal Ritz vectors calculated with the algorithm
of Table 1 based on matrices M
`
i
|, K
`
i
| and F
`
i
| for the
`th substructure. Matrix F
`
i
| consists of only two spa-
tial load vectors (k = 2 in Table 1). The generalized
displacement vector u
`
is much smaller in size than
U
`
i
. Matrix T
`
| is the static transformation matrix
between the internal displacement vector U
`
i
and the
retained displacement vector U
r
.
Based on Eq. (4), the initial displacement vector U
is transformed as follows:
U = [U[ u =
U
1
i
.
.
.
U
`
i
.
.
.
U
r
_
_
_
_
=
X
1
0
.
.
.
T
1
.
.
.
.
.
.
.
.
.
.
.
.
0 X
`
.
.
.
T
`
.
.
.
.
.
.
.
.
.
.
.
.
0 0
.
.
.
I
_
_
_
_
u
1
.
.
.
u
`
.
.
.
U
r
_
_
_
_
(5)
Application of Eq. (5) transformation to Eq. (1) re-
sults in the following reduced dynamic system:
[M[
u [C[
_
u [K[ u = F (6)
where
[M[ = [U[
T
[M[[U[
=
M
1
i
0 M
1
ir
.
.
.
.
.
.
.
.
.
0 M
`
i
M
`
ir
.
.
.
.
.
.
.
.
.
M
1
ir
T
M
`
ir
T
M
r
_
_
_
_
(7a)
M
`
i
= X
`
T
M
`
i
X
`
(7b)
M
`
ir
= X
`
T
(M
`
i
T
`
M
`
ir
) (7c)
M
r
= M
r
`
(T
`
T
M
`
i
T
`
M
`
T
ir
T
`
T
`
T
M
`
ir
) (7d)
Similar expressions hold for C and K. The reduced
load vector F is given by
F =
F
1
i
F
`
i
F
r
T
(8a)
where
F
`
i
= X
`
F
`
i
(8b)
and
F
r
= F
r
`
T
`
F
`
i
(8c)
More information about the crankshaft dynamic
substructuring process and the load-dependent Ritz
vector transformation can be found in Ref. [23].
Fig. 1. Some denitions on the crankshaft substructuring.
2012 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
3.2. Second level of dynamic substructuring analysis
The reduced displacement vector u in Eq. (5)
contains the retained displacement vector U
r
which
can be of a fairly large size. For this reason, a second
level of substructuring is performed which further re-
duces the number of retained degrees of freedom.
The vector U
r
is partitioned as
1
U
r
=
2
U
i 2
U
r
T
(9)
where the left subscript indicates the level of substruc-
turing analysis.
2
U
r
contains the displacements of
three points at the left end, middle and right end of each
main bearing on the crankshaft centerline.
2
U
i
in-
cludes all the other displacements of the substructure
interfaces which were retained in the rst level of sub-
structuring. Due to the partitioning of Eq. (9), the re-
duced displacement vector u becomes
1
u =
2
u
i 2
U
r
T
(10a)
where
2
u
i
=
1
u
1
1
u
`
2
U
i
T
(10b)
The reduced matrices [M[, [C[ and [K[ and the load
vector F from the rst level of substructuring are
partitioned according to Eqs. (10a) and (10b), and the
process described in the previous paragraph is repeated.
A new Ritz transformation matrix [X] is calculated using
again the algorithm of Table 1, and a new reduced sys-
tem, similar to that of Eq. (6), is formed.
It has been found through extensive applications
of the proposed method to a variety of automotive
crankshafts, that 20 Ritz vectors are enough for both
levels of substructuring. Although the number of Ritz
vectors is arbitrary, 1520 Ritz vectors are usually en-
ough to obtain good accuracy in calculating the dy-
namic response of a crankshaft. A larger number can
be also tried until there is no improvement in the cal-
culated dynamic response.
After the two reductions, the reduced model consists
of approximately 50100 degrees of freedom only. This
translates to substantial computational savings in per-
forming a crankshaft dynamic analysis. The computa-
tional savings become even more important when the
crankshaft dynamic analysis is coupled with the main
bearing hydrodynamics to nd the combined system
response.
The two-level dynamic substructuring analysis and
the algorithm to nd the Ritz vectors (Table 1) have
been implemented in the general purpose program MSC MSC/
NASTRAN NASTRAN [28] using DMAP DMAP ALTERS ALTERS [29]. Due to the
excellent data management capabilities and ecient
matrix operations of MSC MSC/NASTRAN NASTRAN, the potentially
Table 1
A subspace algorithm to calculate load-dependent Ritz vectors
1. Given input matrices
[M[: mass matrix (n n)
[K[: stiffness matrix (n n)
[F [: block of k independent spatial load vectors (n k)
2. Triangularize stiness matrix
[K[ = [L[
T
[D[[L[
3. Solve for initial static block [X
1
]
a. Solve for block [x
+
[ = [x
+
1
; x
+
2
; . . . ; x
+
k
[
[K[[x
+
[ = [F [
b. Get block [X[ = [x
1
; x
2
; . . . ; x
k
[ by M-normalization of vectors x
+
j
, j = 1; 2; . . . ; k
x
+
j
T
[M[x
+
j
= a
x
j
= a
1=2
x
+
j
c. Get block [
^
X[ = [^ x
1
; ^ x
2
; . . . ; ^ x
k
[ by M-orthogonalization of vectors in [X[.
Repeat for j = 2; 3; . . . ; k:
[C
j
[ = [x
1
; x
2
; . . . ; x
j1
[
T
[M[x
j
^ x
j
= x
j
[x
1
; x
2
; . . . ; x
j1
[[C
j
[
d. M-normalize block [
^
X[
[
^
X[
T
[M[[
^
X[ = [K[
[X
1
[ = [
^
X[[K[
1=2
4. Solve for subsequent blocks [ X
i
], i = 2; 3; . . . ; p
a. Solve for block [X
+
i
[
[K[[X
+
i
[ = [M[[X
i1
[
b. Get block [
X
i
[ by M-orthogonalization of [X
+
i
[ against all previous blocks
[C[ = [X
1
[; . . . ; [X
i1
[ [ [
T
[M[[X
+
i
[
[
X
i
[ = [X
+
i
[ [X
1
[; . . . ; [X
i1
[ [ [[C[
c. Get block [X
i
[ by M-orthogonalization of [
X
i
[ against all previous blocks according to Step 4b.
d. Get block [X
i
[ by M-orthogonalization of block [X
i
[ according to Steps 3c and 3d.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2013
very large stiness and mass matrices as well as a variety
of other auxiliary large matrices are handled very e-
ciently. Initially, MSC MSC/NASTRAN NASTRAN calculates the mass and
stiness matrices of the model. Then two separate DMAP DMAP
ALTERS ALTERS, instruct MSC MSC/NASTRAN NASTRAN to implement the Ritz
vector algorithm of Table 1 for the two levels of sub-
structuring. In the process, the DMAP DMAP ALTERS ALTERS perform
a number of data manipulation tasks such as partition
and multiplication of large matrices, assembling of data
from dierent substructures, solution of linear systems
of equations, etc. This not only solves the problem of
managing very large matrices but also reduces the overall
computational eort since MSC MSC/NASTRAN NASTRAN performs the
bulk of the computations very eciently.
3.3. Crankshaft loading
The crankshaft is mainly loaded by the engine op-
erating load which comes from the cylinder combustion.
This load is transmitted through the piston and con-
necting rod to the crankpin of the crankshaft. The piston
and the connecting rod are treated as rigid bodies and
their inertia loads are calculated and combined with the
combustion load. The combined load is applied on the
crankpin in the rotating coordinate system. The belt
loads are also applied on the crankshaft.
Since the analysis is performed in the rotating coor-
dinate system, the crankshaft inertia load due to cen-
trifugal forces is also applied on the crankshaft. For a
particular nite-element model of the crankshaft, this
inertia load is equal to mrx
2
, where m is the mass of the
nite element, r the distance of its center of gravity from
the crankshaft rotation axis and x the crankshaft rota-
tional velocity. This load is commonly known as ``WR
segment'' loading. The coriolis forces are neglected.
3.4. Structural damping
Rayleigh damping is used to represent the crankshaft
structural damping. The damping matrix [C[ in Eq. (6) is
taken as
[C[ = a[K[ b[M[ (11)
where a and b are two constants to be determined from
two specied damping ratios which correspond to two
unequal natural frequencies of the system. It has been
shown [30] that if the system eigenvectors are [C]-
orthogonal, the following equation holds
b ax
2
i
= 21
i
x
i
(12)
where x
i
is the ith natural frequency of the system and f
i
is the damping ratio for the ith mode. Given the natural
frequencies and damping ratios of two dierent modes,
the coecients a and b can be determined by solving Eq.
(12). Subsequently, Eq. (12) can be used to calculate f
i
at
any frequency x
i
. The Rayleigh damping is a compu-
tationally convenient way to approximate the actual
structural damping.
3.5. Crankshaft bent and engine block misboring
Crankshaft ``bent'' (Fig. 2a) and engine block mis-
boring (Fig. 2b) represent manufacturing imperfections.
Fig. 2. (a) Schematic of crankshaft ``bent''. (b) Schematic of engine block misboring.
2014 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
The forging, heat treating and machining operations
necessary to produce a crankshaft, sometimes produce a
slightly bent crankshaft. Since this eect is expected, the
crankshaft is subsequently straightened so that it con-
forms to certain nished tolerances. However, it is very
dicult to perfectly straighten the crankshaft. Similarly,
the main bearing bores in the block structure may lie on
dierent centerlines due to slight misboring. Since the
crankshaft bent and the engine block misboring can
induce signicant loading on the main bearings, they are
both included in the formulation of the crankshaft
engine block system model.
4. Representation of engine block
The crankshaft is supported by the engine block at the
main bearing locations. In this work the engine block is
represented by its static stiness neglecting therefore the
engine block dynamic behavior. The engine block sti-
ness can be represented by a vertical and horizontal
value for each main bearing. In practice, the vertical and
horizontal stinesses are always dierent and they both
vary around the circumference of each main bearing.
The engine block support can be modeled by a con-
centrated stiness at mid-bearing position or by a dis-
tributed stiness along the bearing length. In the former
case, the bearing loads consist of reaction forces only. If
these loads are used in a traditional main bearing lu-
brication analysis, there will be no journal misalignment.
However, dynamic misalignment can have a signicant
inuence on the main bearing performance. The issue
of dynamic misalignment can be of concern with the
increasing number of four-cylinder long stroke engines
and V6 congurations where adequate main bearing
performance remains a design challenge.
The journal misalignment can be calculated by
modeling the engine block exibility with a distributed
stiness along the bearing length (Fig. 3a). In this study,
three linear springs are used along the bearing length
(one spring at each bearing end and one spring at the
mid-bearing location) to represent the distributed engine
block exibility in both the vertical and horizontal
planes (Fig. 3b). This representation accounts for the
translational and rotational engine block stiness. As a
result, the bearing loads consist of reaction forces and
reaction moments. If the translational bearing stiness is
k, the three springs have stinesses k=6, 2k=3 and k=6
(Fig. 3b). These stinesses have been calculated so that
the translational and the rotational bearing stinesses
with respect to the mid-bearing location are the same
between the distributed block stiness and the equiva-
lent distributed block stiness models.
At each time step during the dynamic simulation of
the crankshaft motion, the reaction of each main bear-
ing depends on the local engine block stiness which, in
turn, depends on the circumferential location of the re-
sultant main bearing reaction. Therefore, a nonlinear
solver must iterate between assumed local engine block
stinesses for each bearing and the resulting main
bearing reactions until it calculates the latter with rea-
sonable accuracy. For computational eciency, the main
bearing reaction from the previous time step is used in
this work to estimate the local engine block stiness
needed at the current time step. This feature constitutes
a unique engine block stiness representation which
fully accounts for the anisotropy of the engine block
exibility as seen by a rotating crankshaft. The cou-
pling stiness between the vertical and horizontal planes
of the same bearing has been neglected. The coupling
stiness among dierent bearings has also been ne-
glected.
5. Main bearing hydrodynamic analysis
The crankshaft structure is supported by the oil-lm
hydrodynamic pressure of the main bearings. The oil-
lm pressure distribution is described by the Reynolds
equation which is derived from the NavierStokes
equations and the continuity equation under simplifying
assumptions [31]. For a Newtonian uid, the Reynolds
equation is written as
o
oz
h
3
op
oz
_ _
1
R
2
o
oh
h
3
op
oh
_ _
= 6lx
oh
oh
12l
oh
ot
(13)
where p(z; h; t) is the oil-lm pressure, h(z; h; t) is the oil-
lm thickness, l is the oil viscosity, R is the journal
radius and x is the crankshaft angular velocity. The
bearing coordinate system and notation are shown in
Fig. 4. The rst and second terms of the right-hand side
of Eq. (13) represent the shear or wedge lm eect and
the oil squeeze lm eect, respectively. Fig. 3. Representation of the engine block stiness.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2015
There is a prescribed pressure p
h
along the oil grooves
and a prescribed pressure p
a
at the two bearing ends
p(z; h; t) =
p
h
on C
1
p
a
on C
2
_
(14)
where C
1
and C
2
denote the oil-lm domains with
specied pressure. The oil-lm pressure distribution
can be obtained by solving the Reynolds equation (13)
subjected to the boundary conditions of Eq. (14). The
solution however will give negative pressure in the cav-
itated region downstream circumferentially from the line
of minimum lm thickness h
min
(see Fig. 4). For this
reason, a cavitation boundary condition must be used in
solving Eq. (13) in order to obtain a meaningful oil-lm
pressure distribution.
There are many cavitation conditions used in the
literature. The simplest is the Half Sommerfeld cavita-
tion condition in which positive pressures extend
through half the bearing circumference and the other
half is regarded as cavitated [31]. Another cavitation
condition is the G umbel condition in which the Rey-
nolds equation (13) is rst solved subjected to the speci-
ed pressures of Eq. (14) and then the cavitation region
is determined by disallowing the existence of subambient
pressures. A more accurate cavitation condition is the
Reynolds or SwiftStieber condition which requires the
pressure gradient to be zero and the pressure to be equal
to the cavitation pressure at the cavitation boundary.
This condition ensures mass ow continuity at the rup-
ture boundary but not at the reformation boundary
[32,33]. It is adequate for determining load capacity, or
alternatively, minimum lm thickness and journal orbit
under dynamic loading provided that the oil lm is
isothermal and there is oil availability [33]. The most
accurate cavitation condition is the Jakobsson-Floberg
and Olsson (JFO) condition which conserves mass both
at the rupture and the reformation boundaries [32,33].
Comparisons between the Reynolds and JFO conditions
have shown that the former is adequate for bearing load
capacity calculations while the latter is needed to predict
accurately the oil ow in the bearing [33].
Although the Reynolds equation (13) is linear, when
the Reynolds or JFO cavitation conditions are used, the
problem of determining the oil pressure distribution
becomes nonlinear since the cavitation region is not
known a priori. This can increase the computational
cost substantially. It is estimated [34] that in a dynami-
cally loaded bearing the use of the Reynolds cavitation
condition may increase the computational cost by at
least 50%. The G umbel cavitation condition relaxes the
zero-pressure gradient requirement and is therefore, a
compromise to the Reynolds condition. Although it is
unacceptable for oil ow prediction, it introduces an
error of only about 5% in calculating the bearing load
capacity [34,35]. In the present crankshaft system model,
the oil-lm hydrodynamic pressure couples the crank-
shaft dynamic motion with the engine block static mo-
tion. The oil-lm hydrodynamic pressure depends on the
oil-lm thickness which is dened by the crankshaft
orbit within the bearing. Thus, a relatively accurate cal-
culation of the oil-lm pressure and the journal orbit is
needed. On the other hand for practical purposes, the
computational cost in performing a coupled analysis
between the crankshaft dynamics and oil-lm hydrody-
namics must be kept low. Based on this reasoning, the
G umbel cavitation condition is used in the present work
since it provides acceptable accuracy in oil-lm pressure
calculations with minimum computational cost.
5.1. Linear perturbation approach
At a particular time (or crankangle) during the en-
gine cycle, the relative position of the journal within the
bearing denes the oil-lm thickness h. Based on the
oil-lm thickness, the corresponding lm pressure p is
calculated by solving the Reynolds equation (13). Sub-
sequently, p is integrated around the journal to get the
bearing load capacity. In this work, the so-called inverse
lubrication problem must be solved repeatedly as it will
be illustrated later in this report. At time t, the lm
thickness h(z; h; t) is calculated in an iterative manner
until the corresponding load capacity matches a given
applied load.
If the calculation time step Dt is relatively small, both
the lm thickness h(z; h; t) and the corresponding oil-lm
pressure distribution p(z; h; t), or equivalently the load
capacity, will dier from the previous time t Dt by a
small amount Dh and Dp, respectively. This constitutes
the basis of a linear perturbation approach to solve the
Reynolds equation (13). A similar approach has been
Fig. 4. Bearing coordinate system and notation.
2016 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
used in Ref. [36] to calculate the dynamic coecients of
a journal bearing.
At a particular time t, the following assumption is
made
h(z; h; t) = h
0
Dh; and
p(z; h; t) = p
0
Dp
(15)
where h
0
= h(z; h; t Dt) and p
0
= p(z; h; t Dt). Sub-
stitution of Eq. (15) into Eq. (13), retention of only the
rst order terms and separation of variables yields the
following two equations
o
oz
h
3
0
op
0
oz
_ _
1
R
2
o
oh
h
3
0
op
0
oh
_ _
= 6lx
oh
0
oh
12l
oh
0
ot
(16a)
o
oz
h
3
0
oDp
oz
_ _
1
R
2
o
oh
h
3
0
oDp
oh
_ _
=
o
oz
3h
2
0
Dh
op
0
oz
_ _
1
R
2
o
oh
3h
2
0
Dh
op
0
oh
_ _
6lx
oDh
oh
12l
oDh
ot
(16b)
Both Eqs. (16a) and (16b) will be solved using the nite-
element method. The weak (or variational) formulation
and the discretization of the oil-lm domain using a
bilinear rectangular element are described in the fol-
lowing sections.
5.2. Weak formulation
The Galerkin method or weighted residuals method
is used to obtain the variational, or weak, form of Eqs.
(16a) and (16b). Eq. (16a) is multiplied by a virtual
nominal pressure p
0
and Eq. (16b) is multiplied by a
virtual perturbation pressure D p. Subsequently both
equations are integrated over the oil-lm domain X.
Integration by parts yields the following weak forms for
Eqs. (16a) and (16b), respectively
_
X
h
3
0
op
0
oz
o p
0
oz
dX
1
R
2
_
X
h
3
0
op
0
oh
o p
0
oh
dX
=
_
X
6lxh
0
o p
0
oh
dX
_
X
12l
oh
0
ot
p
0
dX (17a)
_
X
h
3
0
oDp
oz
oD p
oz
dX
1
R
2
_
X
h
3
0
oDp
oh
oD p
oh
dX
=
_
X
3h
2
0
op
0
oz
Dh
oD p
oz
dX
1
R
2
_
X
3h
2
0
op
0
oh
Dh
oD p
oh
dX
_
X
6lxDh
oD p
oh
dX
_
X
12l
oDh
ot
D pdX
(17b)
All the boundary terms from the integration by parts are
zero since the pressure is specied on the boundaries
according to Eq. (14).
5.3. Discretization of oil-lm domain using a bilinear
rectangular element
The discretization of the two-dimensional oil-lm
domain is commonly done using linear triangular ele-
ments [37,38]. In this work however, a bilinear rectan-
gular element with ``hourglass control'' is adopted due
to its superior computational eciency without loss of
accuracy.
The ``hourglass control'' technique was rst intro-
duced in Ref. [39]. It is a reduced integration technique
used to form element stiness matrices with a small
computational eort without great loss of accuracy in
the nite-element approximations. Reduced integration
techniques, using lower quadrature rules, are known to
be computationally ecient. However, they result in a
singular global stiness matrix for certain boundary
conditions due to the existence of hourglass (zero en-
ergy) modes. To avoid the singularity, one can use either
a higher quadrature rule in at least one element, or pro-
vide external restraints. The hourglass control method
with a higher quadrature rule can be utilized to eliminate
the singularity [39,40]. The method was rst applied to
lubrication problems in Ref. [41].
The complete derivation of the bilinear rectangular
oil-lm element with hourglass control is presented in
Ref. [42]. The element uidity matrix and the load vector
are derived for an oil-lm element. The two-dimensional
oil-lm domain is discretized using a uniform grid of
rectangular elements. Each rectangular element has di-
mensions (Dz; Dh). After the calculation of the integrals
in the weak formulations, the element equations for the
nominal and perturbation pressure respectively are as
follows [42]:
K
e
|p
0
e
= f
e
0
(18a)
K
e
|Dp
e
= f
e
1
(18b)
where the element uidity matrix [K
e
[ and the element
load vectors f
e
0
and f
e
1
are
[K
e
[ =
(h
e
0
)
3
R
2
Dz
4Dh
1
3
4 2 2 4
2 4 4 2
2 4 4 2
4 2 2 4
_
_
_
_
(h
e
0
)
3
Dh
4Dz
1
3
4 4 2 2
4 4 2 2
2 2 4 4
2 2 4 4
_
_
_
_
(19)
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2017
f
e
0
= 3lxh
e
0
Dz
1
1
1
1
_
_
_
_
3l
oh
0
ot
_ _
e
DhDz
1
1
1
1
_
_
_
_
(20a)
f
e
1
= 6lx
_
3(h
e
0
)
2
R
2
op
0
oh
_
Dh
e
Dz
2
1
1
1
1
_
_
_
_
3(h
e
0
)
2
op
0
oz
Dh
e
Dh
2
1
1
1
1
_
_
_
_
3l
o(Dh
e
)
ot
DhDz
1
1
1
1
_
_
_
_
(20b)
Notice that the uidity matrix for each element can
be calculated by simply multiplying two constant ma-
trices which are common for all elements, by the cubic
power of an average element lm thickness (h
e
0
)
3
. Simi-
larly the load vectors for each element (Eqs. (20a) and
(20b)) are calculated by multiplying constant vectors,
again common among all elements, by the average ele-
ment lm thickness or the average element squeeze lm
rate. The calculation and assembly time for the global
uidity matrix is, therefore, greatly shortened. The re-
sulting computational savings compared to the tradi-
tional linear triangular element used in Ref. [38] is about
40% [41]. In addition, due to the linear perturbation
approach, the uidity matrix [K
e
] is common in Eqs.
(18a) and (18b). Thus it has to be calculated only once at
each time step resulting in substantial computational
savings. Finally, the element equations (18a) and (18b)
are assembled over all elements of the oil domain to get
the corresponding global equations:
[K[p
0
= f
0
(21a)
and
[K[Dp = f
1
(21b)
At each time step, Eq. (21a) is solved once for the
nominal lm pressure p
0
and Eq. (21b) is solved re-
peatedly for Dp by varying Dh, until the total pressure
p = p
0
Dp supports the applied load (inverse lubrica-
tion problem). A modied version of Newton's method
is used for this nonlinear iteration to calculate the cor-
rect Dh at each time step.
Since the lubrication model in this work is coupled
with the crankshaft and engine block structural models,
its computational eciency (without compromising the
accuracy) allows the system model to be a useful prac-
tical design tool. The combination of the linear pertur-
bation approach and the hourglass control technique
ensure the computational eciency of the lubrication
analysis with minimal loss of accuracy. The accuracy
and eciency of the lubrication model is demonstrated
in Ref. [42].
6. Coupled crankshaftengine block model
The reduced crankshaft model is described in the
rotating coordinate system by
[M
c
[u
c
[C
c
[_ u
c
[K
c
[u
c
= F
c
(22)
The vector u
c
is partitioned as
u
c
= u
c
i
[[U
c
r
T
(23)
where u
c
i
is the generalized displacement vector after
the second level of substructuring, U
c
r
is the crank-
shaft retained (subscript r) displacement vector, and
F
c
is the vector of applied forces on the crankshaft.
The superscript c, in the above quantities, denotes the
crankshaft.
The engine block model is described in the rotating
coordinate system by
[C
b
[
_
U
b
r
[K
b
[U
b
r
= F
b
(24)
where U
b
r
is the block retained (subscript r) displace-
ment vector, and F
b
is the vector of applied forces on
the block. The superscript b in the above quantities,
denotes the block.
The mass inertia eect is neglected for the block
model. The block stiness matrix [K
b
[ is given in the
rotating coordinate system as follows
[K
b
[ = [T[
T
k
X
0
0 k
Y
_ _
[T[ =
k
x
k
xy
k
yx
k
y
_ _
(25)
where
[T[ =
cos h sinh
sinh cos h
_ _
(26)
and
k
x
= k
X
cos
2
h k
Y
sin
2
h (27a)
k
y
= k
X
sin
2
h k
Y
cos
2
h (27b)
k
xy
= k
yx
= (k
Y
k
X
) sinh cos h (27c)
The angle of rotation h is measured clockwise from the
vertical position. The block damping matrix [C
b
[ is de-
ned similarly. Inclusion of the block damping is nec-
essary because it dampens any crankshaft free-body
motion.
Let
U
c
T
= U
c
o
U
b
r
(28a)
U
b
T
= U
b
o
U
b
r
(28b)
2018 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
U
c
T
, U
b
T
are the total (subscript T) displacements of the
crankshaft and block respectively, relative to a straight
reference line at the mains (Fig. 2), including elastic
deformation.
U
c
o
, U
b
o
are the crankshaft ``bent'' and engine block
misboring, respectively. They are known deviations or
osets (subscript o) from the same reference line.
U
c
r
, U
b
r
are the crankshaft and block retained dis-
placement vectors, respectively. They represent the
elastic deformation of the crankshaft and block from
their undeformed position.
All the above vectors include the crankshaft and
block displacements at the left end, middle position and
right end of each main bearing. The reference line nee-
ded to dene the vectors is arbitrarily taken as a straight
line passing through the middles of the rst (fan end)
and the last (ywheel) bearings. The osets of the rst
and last bearings are arbitrarily set to zero. The ``run-
out'' of the other bearings, or the amount each bearing
deviates from that straight line in the rotating crankshaft
coordinate system, is then entered as oset. The engine
block misboring is specied in a similar way but in the
xed coordinate system.
Based on the partitioning of Eq. (23), Eq. (22) can be
written as
M
c
ii
M
c
ir
M
c
T
ir
M
c
rr
_ _
u
c
i
U
c
r
_ _
C
c
ii
C
c
ir
C
c
T
ir
C
c
rr
_ _
_ u
c
i
_
U
c
r
_ _
K
c
ii
K
c
ir
K
c
T
ir
K
c
rr
_ _
u
c
i
U
c
r
_ _
=
F
c
i
F
c
r
_ _
(29)
The bearing eccentricity is dened as the dierence of
the total displacements of the crankshaft and block
at the bearing locations; i.e.
e = U
c
T
U
b
T
(30)
or due to Eqs. (28a) and (28b)
U
b
r
= U
c
o
U
b
o
U
c
r
e (31)
Also due to action and reaction,
F
c
r
= F
b
(32)
Combining Eqs. (24), (31) and (32) yields
F
c
r
= [C
b
[(_ e
_
U
b
o
_
U
c
o
) [K
b
[(e
U
b
o
U
c
o
) [C
b
[
_
U
c
r
[K
b
[U
c
r
(33)
Since the crankshaft ``bent'' U
c
o
is a vector of con-
stants in the rotating coordinate system,
_
U
c
o
is always
equal to zero. However, since the block misboring U
b
o
C
c
ii
C
c
ir
C
c
T
ir
C
c
rr
C
b
_ _
_ u
c
i
_
U
c
r
_ _
K
c
ii
K
c
ir
K
c
T
ir
K
c
rr
K
b
_ _
u
c
i
U
c
r
_ _
=
F
c
i
F
+
_ _
(34a)
where
F
+
= [C
b
[(_ e
_
U
b
o
) [K
b
[(e U
b
o
U
c
o
)
(34b)
If the bearing hydrodynamics are neglected, e = 0.
However even in this case, the crankshaft bent and block
misboring produce equivalent forces according to Eq.
(34b), which are applied on the crankshaft. All quanti-
ties in Eqs. (34a) and (34b) must be expressed in the
rotating coordinate system.
Before Eq. (34a) are integrated in time to nd the
response of the reduced system, a torsional boundary
condition is applied at the ywheel end of the crankshaft
to eliminate the rotational free-body motion. For ma-
nual transmission applications or dynamometer tests,
the torsional boundary condition consists of a torsional
spring representing the propshaft torsional stiness. For
automatic transmission applications, a torsional damper
boundary condition is used in order to capture the slip-
ping of the torque converter.
Eqs. (34a) and (34b) are nonlinear since the nonlinear
bearing hydrodynamics are implicitly involved through
the bearing eccentricity vector e. A modied New-
mark method [43] with a NewtonRaphson based it-
eration at each time step, is employed for the time
integration of Eqs. (34a) and (34b). At each time step,
the following iteration loop is considered:
Step 1: Assume the bearing eccentricity vector e.
Step 2: Calculate the bearing reaction forces and
moments F
b
HYDRO
from the bearing hydrodynamics
algorithm, using the assumed vector e from Step 1.
Step 3: Calculate the crankshaft retained displace-
ment vector U
c
r
from Eqs. (34a) and (34b), based on
the assumed e.
Step 4: Calculate the block retained displacement
vector U
b
r
from Eq. (31).
Step 5: Calculate the bearing reaction forces and
moments F
b
from Eq. (24), based on the calculated
U
b
r
from Step 4.
Step 6: If the absolute dierence between F
b
and
F
b
HYDRO
is greater than a small tolerance, update e
and repeat the process from Step 1. Otherwise, proceed
to the next time step.
The solution of Eqs. (34a) and (34b) gives the dy-
namic displacements of the reduced crankshaftengine
block system model. The displacements of the original
crankshaft model can be obtained by two backwards
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2019
transformations (one for each level of substructuring).
The displacements of the original model can be subse-
quently processed to calculate the crankshaft operating
dynamic stresses.
7. Selected results
7.1. Dynamic analysis of an in-line ve-cylinder engine
In order to experimentally validate the accuracy of
the proposed methodology, the dynamic analysis of a
ve-cylinder in-line (L5) engine crankshaft was per-
formed using the nite-element based analytical tool
CRANKSYM which implements the methodology pre-
sented in this paper. The response of the L5 crankshaft
under wide-open-throttle (WOT) operating conditions
was simulated and correlated with measurements.
Figs. 5 and 6 show the measured and computed an-
gular vibrations of the L5 crankshaft, respectively in
``waterfall'' format. The torsional vibration amplitude in
degrees is plotted for dierent engine orders and engine
speeds. Note that the measured and simulated vibration
amplitudes are peak-to-peak and single amplitude (half
of peak-to-peak), respectively. There is a very good
correlation between the simulated with CRANKSYM
and the measured vibrations at all engine speeds except
around 6200 rpm at 2.5 engine order where CRANK-
SYM's prediction is about 20% higher than the mea-
surement. The dierence is mainly due to two reasons.
First, the inherent nonlinearity of the tuned torsional
vibration absorber's elastomer and second the coupling
between the absorber's inertia ring and the accessory
drive components.
The behavior of the nonlinear tuned absorber is such
that higher vibration amplitudes would decrease the
stiness and increase the damping of the elastomer. This
change in the elastomer material properties with defor-
mation results in a decrease of the crankshaft resonance
frequency and an increase of the crankshaft dam-
ping. The conducted simulations did not consider the
nonlinear eects of the tuned absorber elastomer and
Fig. 5. Measured vibration for an in-line, ve-cylinder engine.
Fig. 6. Simulated vibration for an in-line, ve-cylinder engine.
Fig. 7. Measured vibration for an in-line, ve-cylinder engine
with locked absorber's ring.
Fig. 8. Simulated vibration for an in-line, ve-cylinder engine
with locked absorber's ring.
2020 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
Fig. 9. The nite-element mesh for the pulleycrankshaftexplate system of the PV6 engine.
Fig. 10. Dynamic vs static calculation of bearing loads and moments.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2021
therefore, such small dierences between simulated and
measured vibrations were expected.
The accessory drive components were found to
interact with the tuned absorber's inertia ring since the
latter is used as a pulley. This interaction seems to
manifest itself as a decrease in the inertia of the ring
which eectively increases the tuning frequency of the
absorber. This inertia reduction eect was calculated
and included in the simulations by decreasing the ring's
inertia and increasing the tuning frequency of the ab-
sorber appropriately. The eect of the accessory drive
components on the eective inertia of the absorber's ring
can be signicant and depends on which accessory com-
ponents are present and whether A/C is engaged.
In order to eliminate the inherent nonlinearity of the
tuned torsional vibration absorber's elastomer, the ab-
sorber's ring was pinned to the hub (locked ring) deac-
tivating therefore, the eect of the elastomer's damping.
Furthermore, the coupling between the absorber's iner-
tia ring with the accessory drive components was elimi-
nated by disconnecting the accessory drives from the
crank pulley in a control dynamometer test. Figs. 7 and
8 show the measured and computed angular vibrations
of the L5 crankshaft, respectively in this case. The
agreement now is excellent. Note also that the peaks of
the resonances are more pronounced since the ab-
sorber's damping is eliminated.
7.2. Dynamic analysis of a V6 (six-cylinder) engine
The V6 engine crankshaft, shown in Fig. 9, is used to
demonstrate the capabilities of the proposed methodol-
ogy and the importance of various design parameters in
a crankshaftblock interaction study. Both the crank-
shaft pulley and the explate are included in the model.
The nite-element mesh consists of 6814 solid elements
and 8962 grids (26,886 degrees of freedom). The analysis
was performed at 4500 rpm. Twenty Ritz vectors were
used for both the rst and the second substructuring.
Both the combustion load and the inertia load due to
reciprocating and rotating masses of the crankcon-
necting rodpiston assembly were considered in the
calculation of the crankpin loads.
The eect of calculating bearing loads using a dy-
namic analysis as opposed to a static analysis is shown in
Fig. 10. The vertical and horizontal loads as well as the
vertical and horizontal bending moments are compared
for the rst bearing from the pulley end. A constant
block stiness of 100,000 N/mm in both the vertical
and horizontal planes is used. As shown, the maximum
Fig. 11. Eect of block stiness on bearing loads and moments.
2022 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
dynamic load is considerably higher than the maximum
static load in both the vertical and horizontal planes.
Substantial dierences also exist in the vertical and
horizontal bending moments. A large bending moment
may generate a large journal misalignment which in
turn, may alter the bearing hydrodynamic performance
substantially.
Fig. 11 illustrates the eect of the block stiness on
bearing loads. The rigid, exible and soft cases are
compared for the second bearing. A 100,000 N/mm
block stiness is used for the exible case which is rep-
resentative of iron blocks. The block stiness for the
rigid and soft cases are 600,000 and 25,000 N/mm which
are articially large and small for the two respective
cases. All stinesses were assumed circumferentially
uniform for all bearings. The dierence in both loads
and moments, between the rigid and soft cases are
substantial. This indicates the importance of the block
stiness in crankshaft dynamic analysis and clearly
suggests the importance of the block dynamic behavior
(not studied here).
The inuence of engine block stiness distribution
around the bearing bore on the vertical and horizontal
loads and moments is shown in Fig. 12 for the second
bearing. The solid lines correspond to a uniform (not
varying circumferentially) block stiness of 100,000 N/
mm. For the dashed lines, the block stiness varies lin-
early between the values of 150,000 and 50,000 N/mm at
the top (bulkhead) and bottom (bearing cap) locations
of the bearing bore, respectively. The average of these
two values is 100,000 N/mm which is representative of
iron blocks. As shown in the gure, the circumferentially
varying stiness results in a maximum vertical load
reduction of approximately 3000 N and a maximum
horizontal moment reduction of approximately 4000
Nmm.
Fig. 13 shows the eect of block misboring on the
loads and moments of the fourth bearing. The misboring
at each bearing location is arbitrarily dened by moving
the third bearing upwards by 0.1 mm. All the other
bearings are assumed to lie on a straight line. A constant
block stiness of 100,000 N/mm is used for all the
bearings. The misboring distribution does not practi-
cally change the vertical and horizontal bearing loads.
However, it drastically changes the bending moments.
This suggests that the location of the bearing resultant
loads is substantially moved along the bearing length.
As a result, bearing end loading can occur with severe
lubrication consequences and possible bearing fail-
ure. For this reason, any bearing misboring due to
Fig. 12. Eect of circumferential block stiness distribution on bearing loads and moments.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2023
manufacturing imperfections should always be consid-
ered. Although results for the crankshaft bent are not
presented here, it has been found having a similar to the
block misboring eect.
Fig. 13. Eect of block misboring on bearing loads and moments.
Fig. 14. Eect of bearing hydrodynamics on bearing loads and moments.
2024 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
Fig. 15. Calculated journal misalignment in the vertical plane under dynamic loading.
Fig. 16. Calculated bearing minimum lm thickness under dynamic loading.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2025
The eect of bearing hydrodynamics on the loads and
moments of the second bearing is shown in Fig. 14. A
25 lm clearance is used for all bearings. A constant
block stiness of 100,000 N/mm is assumed. As shown in
the gure, the inclusion of the bearing hydrodynamics
in the analysis increased the maximum vertical load and
the maximum horizontal moment considerably. Fig. 15
shows the vertical eccentricity for all four bearings at the
left and right ends of the bearing. The dierence between
the left and right eccentricity indicates the amount of
misalignment within each bearing. During most of the
engine cycle, the bearing misalignment is considerable
for all bearings. Since the misalignment can inuence the
hydrodynamic behavior of the bearing, it should be
always considered when the bearing performance is as-
sessed. Finally, Fig. 16 shows the minimum lm thick-
ness for all bearings. The minimum lm thickness is
around 5 lm for all bearings except the rst bearing for
which it is around 2 lm. This indicates that the bearings
are not heavily loaded for this operating condition. For
heavily loaded bearings, the minimum lm thickness can
be below 0:5 lm.
Acknowledgements
The author wishes to express his gratitude to Dr.
Turgay Bengisu from the Synthesis and Analysis De-
partment of the General Motors Powertrain Division for
providing access to the experimental results presented in
this paper. Furthermore, his eorts in productionizing
and institutionalizing CRANKSYM within General
Motors are greatly appreciated. His knowledge and ex-
perience on engine crankshaft vibrations and on vibra-
tions in general have been invaluable in developing and
improving CRANKSYM over the past few years.
References
[1] Henry JP, Toplosky J, Abramczuk M. Crankshaft dura-
bility prediction a new 3-D approach. SAE paper no.
920087, 1992.
[2] Heath AR, McNamara PM. Crankshaft stress analysis
the combination of nite element and classical techniques.
ASME ICE 1989;9.
[3] Kubota M, Kamichika R, Tanida K, Nakagawa E.
Dynamic analysis of crankshaft using component mode
synthesis. Part 1: ecient method of calculations utilizing
NASTRAN image superelements. Bull Marine Engng Soc
Jpn 1988;16(2).
[4] Mourelatos ZP. An analytical investigation of the crank-
shaftywheel bending vibrations for a V6 Engine. SAE
International Proceedings of the 1995 Noise and Vibration
Conference, vol. 1, SAE paper no. 951276, 1995.
[5] Kubozuka T, Hayashi Y, Hayakawa Y, Kikuchi K.
Analytical study on engine noise caused by vibration of
the cylinder block and crankshaft. SAE paper no. 830346,
1983.
[6] Ide S, Uchida T, Ozawa K, Izawa K. Improvement of
engine sound quality through a new ywheel system
exibly mounted to the crankshaft. SAE paper no.
900391, 1990.
[7] Ochini K, Nakano M. Relation between crankshaft
torsional vibration and engine noise. SAE paper no.
790365, 1979.
[8] Dejong R. Using vibration transmission analyses in the
design of quiet engines. In: Hickling R, Kamal M, editors.
Engine noise, excitation, vibration and radiation. New
York: Plenum Press; 1982. p. 12346.
[9] Welsh WA. Dynamic analysis of engine bearing systems.
MS Thesis, Cornell University, 1982.
[10] Katano H, Iwamoto A, Saitoh T. Dynamic behaviour
analysis of internal combustion engine crankshafts under
operating conditions. Institute of Mechanical Engineers,
Paper no. C430/049, 1991. p. 20516.
[11] Nefske DJ, Sung SH. Coupled vibration response of the
engine crankblock system. ASME 1989;DE-18-4:37985.
[12] Mayer LS, Zeischka J, Scherens M, Maessen F. Analysis of
exible rotating crankshaft with exible engine block using
MSC MSC/NASTRAN NASTRAN and DADS DADS. 1995 MSC World Users'
Conference, 1995.
[13] Martin IT, Law B. Prediction of crankshaft and ywheel
dynamics. Institute of Mechanical Engineers, Paper no.
C382/046, 1989.
[14] Law B, Haddock AK. Prediction of main bearing and
crankshaft loading in reciprocating engines. 15th CIMAC
Congress, Paris, 1983.
[15] Okamura H, Shinno A, Yamanaka T, Suzuki A, Sogabe
K. A dynamic stiness matrix approach to the analysis
of three-dimensional vibrations of automobile engine
crankshafts. Part I: background and application to free
vibrations. ASME Winter Annual Meeting, Dallas, TX,
1990.
[16] Morita T, Okamura H. A dynamic stiness matrix
approach to the analysis of three-dimensional vibrations
of automobile engine crankshafts. Part 2: application to
ring conditions. ASME Winter Annual Meeting, Dallas,
TX, 1990.
[17] Okamura H, Morita T. Inuence of crankshaftpulley
dimensions on crankshaft vibrations and engine-structure
noise and vibrations. SAE International Proceedings of the
1993 Noise and Vibration Conference. SAE paper no.
931303, 1993. p. 32938.
[18] Stickler AC. Calculation of bearing performance in inde-
terminate systems. PhD Thesis, Cornell University, 1974.
[19] Booker JF. Dynamically loaded journal bearings: numer-
ical application of the mobility method. Trans ASME,
J Lubricat Technol 1971;93(1).
[20] Paranjpe RS, Cusenza A. FLARE FLARE: an integrated software
package for friction and lubrication analysis of automotive
engines. Part II: experimental validation. SAE Publication
SP-919 on Engine Tribology, SAE paper no. 920488, 1992.
p. 5765.
[21] Goenka PK, Paranjpe RS. A review of engine bearing
analysis methods at General Motors. SAE Publication SP-
919 on Engine Tribology, SAE paper no. 920489, 1992.
p. 6775.
2026 Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027
[22] Ishihama M, Hayashi Y, Kubozuka T. An analysis of the
movement of the crankshaft journals during engine ring.
SAE paper no. 810772, 1981.
[23] Mourelatos ZP. An ecient crankshaft dynamic analysis
using substructuring with Ritz vectors. J Sound Vib
2000;238(3):495527.
[24] Wilson EL, Yuan MW, Dickens JM. Dynamic analysis by
direct superposition of Ritz vectors. Earthquake Engng
Struct Dynam 1982;10:81321.
[25] Arnold RR, Citerley RL, Chagrin M, Galant D. Applica-
tion of Ritz vectors for dynamic analysis of large
structures. Comput Struct 1985;21:4617.
[26] Wilson EL, Bayo EP. Use of special Ritz vectors in
dynamic substructure analysis. J Struct Engng 1986;112(8):
194454.
[27] Leger P, Wilson EL. Generation of load dependent Ritz
transformation vectors in structural dynamics. Engng
Comput 1987;4:30918.
[28] Gockel MA, editor, MSC MSC/ / NASTRAN NASTRAN: handbook for dy-
namic analysis. The MacNeal Schwendler Corp., 1983.
[29] Reymond MA, editor, MSC MSC/ / NASTRAN NASTRAN: User's manual. The
MacNeal-Schwendler Corp., 1991.
[30] Bathe KJ, Wilson EL. Numerical methods in nite element
analysis. Englewood Clis NJ: Prentice-Hall; 1976.
[31] Pinkus O, Sternlicht B. Theory of hydrodynamic lubrica-
tion. New York: McGraw-Hill; 1961.
[32] Brewe DE. Theoretical modeling of the vapor cavitation
in dynamically loaded journal bearings. Trans ASME,
J Tribol 1986;108:62838.
[33] Paranjpe RS, Goenka PK. Analysis of crankshaft bearings
using a mass conserving algorithm. STLE Tribol Trans
1990;33(3):33344.
[34] Szeri AZ, editor. Tribology, friction, lubrication and wear.
New York: Hemisphere; 1980.
[35] Cameron A. The principles of lubrication. New York:
Wiley; 1966.
[36] Klit P, Lund JW. Calculation of the dynamic coecients of
a journal bearing, using a variational approach. ASME
Paper 85-Trib-7. ASME/ASLE Joint Lubrication Confer-
ence, Atlanta, GA, October 1985.
[37] Goenka PK. Dynamically loaded journal bearings: nite-
element method analysis. Trans ASME, J Tribolo Series F
1984:42939.
[38] Mourelatos ZP, Parsons MG. Finite-element analysis of
elastohydrodynamic stern bearings. SNAME Trans 1985;
93:22559.
[39] Belytschko T, Liu WK, Kennedy JM. Hourglass control in
linear and nonlinear problems. In: Alturi S, Perrone N,
editors. Computer methods for nonlinear solids and
structures, vol. 54, ASME/AMD. American Society of
Mechanical Engineers, New York, 1983. p. 3763.
[40] Kikuchi N. Finite element methods in mechanics. Depart-
ment of Mechanical Engineering, The University of
Michigan, 1985.
[41] Karni ZH, Parsons MG, Mourelatos ZP. Time-varying
behavior of a statically indeterminate shafting system in a
hydrodynamic journal bearing. ASME J Tribol 1987;
109(1).
[42] Mourelatos ZP. An ecient oil lm lubrication analysis for
engine crankshafts. Tribology Transactions 2001;44(3):
35158.
[43] Newmark NM. A method of computation for structural
dynamics. Proceedings of the American Society of Civil
Engineers, 1959.
Z.P. Mourelatos / Computers and Structures 79 (2001) 20092027 2027